North American Network Operators Group

Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical

Re: Formal study: How many points networks share

  • From: Sean Donelan
  • Date: Mon Aug 28 20:26:23 2000

On Mon, 28 August 2000, batz wrote:
> I missed the discussion from a few weeks ago, but is this for physical 
> network maps? 

Yes.

> If not, couldn't this be done using routing tables, RAdb 
> information, and some geometry a la CAIDA? 
>
> Aren't most accurate physical network maps borderline classified for the
> most part? 

Yes, but not a problem for the government.  They've done this type of study
for the voice network, and already have NDA's with the three major IXCs and
however many RBOCs are left.  I doubt getting NDA's from the other major CLECs
is a real issue.  Its mostly deciding which ones actually matter.  I wouldn't
use the word "classified," it has a specialized meaning in this context. The
problem isn't NDAs, but rather it will be too NDA.  Even the contributors may
not find out about their own weaknesses, which may be great for the government
but not much use for industry.

> Many network maps include a PVC they use from a Tier-1 provider as part
> of a 'physical' infrastructure, which would cause their network map to 
> be conspicuously similar to that of their transit provider. 

And they would have shared fate.  However, unless you know how the logical
network is routed across the physical network you can't really predict the
affects of physical actions.  You might think there are two routes which
don't pass through the same city, but in fact the physical layer of the
PVC may not only pass through the same city, but the same switch and same
fiber.

It may turn out the real vulnerability is some bridge across the Mississippi
river in the middle of nowhere which 90% of the carriers use as the route
from east to west across the country.

> CAIDA's AS connectivity maps are the closest thing to a study that I have
> seen.

The CAIDA maps are a good starting point, and in fact have been used as the
starting point.  But they have some severe limitations for predicting the
behavior of the network.  In particular they can't discover "shadow" routes
until after the primary route falls out of service.

While the details of the net may change from microsecond to microsecond, the
macro level stuff changes much more slowly.  Providers aren't opening and
closing POPs minute by minute.  Nor do you need a 100% map to figure out how
you can effectively knock New York City off the net.